1 /* -*- mode: c; c-basic-offset: 8; -*- 2 * vim: noexpandtab sw=8 ts=8 sts=0: 3 * 4 * Copyright (C) 2005 Oracle. All rights reserved. 5 * 6 * This program is free software; you can redistribute it and/or 7 * modify it under the terms of the GNU General Public 8 * License as published by the Free Software Foundation; either 9 * version 2 of the License, or (at your option) any later version. 10 * 11 * This program is distributed in the hope that it will be useful, 12 * but WITHOUT ANY WARRANTY; without even the implied warranty of 13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 14 * General Public License for more details. 15 * 16 * You should have received a copy of the GNU General Public 17 * License along with this program; if not, write to the 18 * Free Software Foundation, Inc., 59 Temple Place - Suite 330, 19 * Boston, MA 021110-1307, USA. 20 */ 21 22 #ifndef O2CLUSTER_TCP_INTERNAL_H 23 #define O2CLUSTER_TCP_INTERNAL_H 24 25 #define O2NET_MSG_MAGIC ((u16)0xfa55) 26 #define O2NET_MSG_STATUS_MAGIC ((u16)0xfa56) 27 #define O2NET_MSG_KEEP_REQ_MAGIC ((u16)0xfa57) 28 #define O2NET_MSG_KEEP_RESP_MAGIC ((u16)0xfa58) 29 30 /* we're delaying our quorum decision so that heartbeat will have timed 31 * out truly dead nodes by the time we come around to making decisions 32 * on their number */ 33 #define O2NET_QUORUM_DELAY_MS ((o2hb_dead_threshold + 2) * O2HB_REGION_TIMEOUT_MS) 34 35 /* 36 * This version number represents quite a lot, unfortunately. It not 37 * only represents the raw network message protocol on the wire but also 38 * locking semantics of the file system using the protocol. It should 39 * be somewhere else, I'm sure, but right now it isn't. 40 * 41 * New in version 7: 42 * - DLM join domain includes the live nodemap 43 * 44 * New in version 6: 45 * - DLM lockres remote refcount fixes. 46 * 47 * New in version 5: 48 * - Network timeout checking protocol 49 * 50 * New in version 4: 51 * - Remove i_generation from lock names for better stat performance. 52 * 53 * New in version 3: 54 * - Replace dentry votes with a cluster lock 55 * 56 * New in version 2: 57 * - full 64 bit i_size in the metadata lock lvbs 58 * - introduction of "rw" lock and pushing meta/data locking down 59 */ 60 #define O2NET_PROTOCOL_VERSION 7ULL 61 struct o2net_handshake { 62 __be64 protocol_version; 63 __be64 connector_id; 64 __be32 o2hb_heartbeat_timeout_ms; 65 __be32 o2net_idle_timeout_ms; 66 __be32 o2net_keepalive_delay_ms; 67 __be32 o2net_reconnect_delay_ms; 68 }; 69 70 struct o2net_node { 71 /* this is never called from int/bh */ 72 spinlock_t nn_lock; 73 74 /* set the moment an sc is allocated and a connect is started */ 75 struct o2net_sock_container *nn_sc; 76 /* _valid is only set after the handshake passes and tx can happen */ 77 unsigned nn_sc_valid:1; 78 /* if this is set tx just returns it */ 79 int nn_persistent_error; 80 81 /* threads waiting for an sc to arrive wait on the wq for generation 82 * to increase. it is increased when a connecting socket succeeds 83 * or fails or when an accepted socket is attached. */ 84 wait_queue_head_t nn_sc_wq; 85 86 struct idr nn_status_idr; 87 struct list_head nn_status_list; 88 89 /* connects are attempted from when heartbeat comes up until either hb 90 * goes down, the node is unconfigured, no connect attempts succeed 91 * before O2NET_CONN_IDLE_DELAY, or a connect succeeds. connect_work 92 * is queued from set_nn_state both from hb up and from itself if a 93 * connect attempt fails and so can be self-arming. shutdown is 94 * careful to first mark the nn such that no connects will be attempted 95 * before canceling delayed connect work and flushing the queue. */ 96 struct delayed_work nn_connect_work; 97 unsigned long nn_last_connect_attempt; 98 99 /* this is queued as nodes come up and is canceled when a connection is 100 * established. this expiring gives up on the node and errors out 101 * transmits */ 102 struct delayed_work nn_connect_expired; 103 104 /* after we give up on a socket we wait a while before deciding 105 * that it is still heartbeating and that we should do some 106 * quorum work */ 107 struct delayed_work nn_still_up; 108 }; 109 110 struct o2net_sock_container { 111 struct kref sc_kref; 112 /* the next two are vaild for the life time of the sc */ 113 struct socket *sc_sock; 114 struct o2nm_node *sc_node; 115 116 /* all of these sc work structs hold refs on the sc while they are 117 * queued. they should not be able to ref a freed sc. the teardown 118 * race is with o2net_wq destruction in o2net_stop_listening() */ 119 120 /* rx and connect work are generated from socket callbacks. sc 121 * shutdown removes the callbacks and then flushes the work queue */ 122 struct work_struct sc_rx_work; 123 struct work_struct sc_connect_work; 124 /* shutdown work is triggered in two ways. the simple way is 125 * for a code path calls ensure_shutdown which gets a lock, removes 126 * the sc from the nn, and queues the work. in this case the 127 * work is single-shot. the work is also queued from a sock 128 * callback, though, and in this case the work will find the sc 129 * still on the nn and will call ensure_shutdown itself.. this 130 * ends up triggering the shutdown work again, though nothing 131 * will be done in that second iteration. so work queue teardown 132 * has to be careful to remove the sc from the nn before waiting 133 * on the work queue so that the shutdown work doesn't remove the 134 * sc and rearm itself. 135 */ 136 struct work_struct sc_shutdown_work; 137 138 struct timer_list sc_idle_timeout; 139 struct delayed_work sc_keepalive_work; 140 141 unsigned sc_handshake_ok:1; 142 143 struct page *sc_page; 144 size_t sc_page_off; 145 146 /* original handlers for the sockets */ 147 void (*sc_state_change)(struct sock *sk); 148 void (*sc_data_ready)(struct sock *sk, int bytes); 149 150 struct timeval sc_tv_timer; 151 struct timeval sc_tv_data_ready; 152 struct timeval sc_tv_advance_start; 153 struct timeval sc_tv_advance_stop; 154 struct timeval sc_tv_func_start; 155 struct timeval sc_tv_func_stop; 156 u32 sc_msg_key; 157 u16 sc_msg_type; 158 159 struct mutex sc_send_lock; 160 }; 161 162 struct o2net_msg_handler { 163 struct rb_node nh_node; 164 u32 nh_max_len; 165 u32 nh_msg_type; 166 u32 nh_key; 167 o2net_msg_handler_func *nh_func; 168 o2net_msg_handler_func *nh_func_data; 169 o2net_post_msg_handler_func 170 *nh_post_func; 171 struct kref nh_kref; 172 struct list_head nh_unregister_item; 173 }; 174 175 enum o2net_system_error { 176 O2NET_ERR_NONE = 0, 177 O2NET_ERR_NO_HNDLR, 178 O2NET_ERR_OVERFLOW, 179 O2NET_ERR_DIED, 180 O2NET_ERR_MAX 181 }; 182 183 struct o2net_status_wait { 184 enum o2net_system_error ns_sys_status; 185 s32 ns_status; 186 int ns_id; 187 wait_queue_head_t ns_wq; 188 struct list_head ns_node_item; 189 }; 190 191 #endif /* O2CLUSTER_TCP_INTERNAL_H */ 192